ISO and noise - different points of view

[Roland Karlsson]

If you set EC to a negative value in order to get a darker picture, then the noise will not be affected.

Oh, but it will.

Yet another proof that you do not read.

Okay, I will put you on my Ignore User list as I find that very offensive.

Read again and come with another answer.

Nonetheless, I will respond yet again here. Yes, I read what you wrote, and I have done the experiment that shows that it is wrong. Your assertion (as I read it before, and as I still read it) is that there is no difference in the noise choosing a higher ISO and lowering brightness during Raw conversion or choosing a lower ISO and raising brightness during Raw conversion, and my answer to that remains the same.

That is not what I wrote above.

I wrote that if you use a negative EC in order to make the image darker, then you do not increase the noise.

And this is true.

 

[Tony Beach]

If you set EC to a negative value in order to get a darker picture, then the noise will not be affected.

Oh, but it will.

Yet another proof that you do not read.

Okay, I will put you on my Ignore User list as I find that very offensive.

Read again and come with another answer.

Nonetheless, I will respond yet again here. Yes, I read what you wrote, and I have done the experiment that shows that it is wrong. Your assertion (as I read it before, and as I still read it) is that there is no difference in the noise choosing a higher ISO and lowering brightness during Raw conversion or choosing a lower ISO and raising brightness during Raw conversion, and my answer to that remains the same.

That is not what I wrote above.

I wrote that if you use a negative EC in order to make the image darker, then you do not increase the noise.

And this is true.

It appears you need to re-read what you wrote right after that:

"But, if you also compensate in your editor to make the picture bright again, then you have increased noise, the same way as increasing ISO does."

My experiment showed that is wrong.

 

[ThrillaMozilla]

* (The actually effect of the ISO setting is camera-specific. Increasing ISO almost always decreases the read noise except in a very few cameras, and it never causes an increase.)

Rarely ever say "never"; there are cameras that have more input-referenced pre-gain read noise with more analog gain, like the Panasonic FZ50.

Bad example. An increase from 6.02 electrons to 6.06 electrons is well within the measurement error. There are better examples of nonmonotonic noise-ISO curves, (i.e. wiggles in the curve), but in the context of noise, the wiggles are usually minor, and the overall trend is almost always neutral or downward. But yes, it was supposed to say "almost never".

http://www.photonstophotos.net/Charts/RN_e.htm#Panasonic Lumix DMC-FZ50_12,Sony DSLR-A850_12

 

[Roland Karlsson]

It appears you need to re-read what you wrote right after that:

"But, if you also compensate in your editor to make the picture bright again, then you have increased noise, the same way as increasing ISO does."

My experiment showed that is wrong.

You need no experiment for that. If you expose less you do get less photons and you get a worse signal to noise ratio.

You can show that by taking it to the limit - expose 10 stops too little and then boost the image 10 stops in your editor.

 

[Roland Karlsson]

The flaw is attributing the noise to the ISO and not to the exposure.

You obviously have not read what I wrote.

If you increase the ISO in auto exposure you will decrease the exposure and therefore increase the noise in the picture.

Not if you correspondingly adjust EC.

I never said that. That is your personal addition.

And BTW - if you do you might over expose the bright parts.

Tadaaaaa!

Whatever.

What are you complaining about and calling flaw?

Straight up wrongheaded thinking. Specifically, "... some otherwise very well-informed and intelligent people still persist in arguing that increased ISO does not cause increased noise!"

 

[Tony Beach]

The good idea is to optimize the exposure and then select the optimal ISO for that exposure; not the other way around.

In a sense, optimizing ISO first is simply impossible. It needs to be optimized depending on the exposure

There is an exception: in an "unlimited light, high dynamic range scene" (like a landscape on a tripod).

In these cases, set the "optimal" ISO first (which will usually be the base ISO), and then set the exposure not to clip important bits at this ISO.

I don't know what "unlimited light" really is. The sun perhaps? Anyway, if I have too much light because I want to use a wide aperture and/or a slow shutter speed then I put an appropriate ND filter on the lens. To say I "chose" that ISO is the same as saying I chose to shoot at 1/8000s or stop down the lens too far -- they are just limitations that I have to deal with.

Also, the opposite is true. If I have too low an EV I have to account for my gear's constraints there too.

 

[Tony Beach]

It appears you need to re-read what you wrote right after that:

"But, if you also compensate in your editor to make the picture bright again, then you have increased noise, the same way as increasing ISO does."

My experiment showed that is wrong.

You need no experiment for that. If you expose less you do get less photons and you get a worse signal to noise ratio.

You entirely missed the point of the experiment. The exposure is the same for both the ISO 100 and ISO 3200 shots, what changed was the SNR which was lower for the ISO 100 shot (thus more noise) than it was for the ISO 3200 shot.

You can show that by taking it to the limit - expose 10 stops too little and then boost the image 10 stops in your editor.

Five was a reasonable enough number.

 

[Tony Beach]

The flaw is attributing the noise to the ISO and not to the exposure.

You obviously have not read what I wrote.

If you increase the ISO in auto exposure you will decrease the exposure and therefore increase the noise in the picture.

Not if you correspondingly adjust EC.

I never said that. That is your personal addition.

You in fact responded to that and wrote it makes no difference.

And BTW - if you do you might over expose the bright parts.

Nonetheless, higher ISO does not necessarily mean higher noise. It only happens when you add numerous qualifications to that statement (e.g., no use of EC or adjusting brightness by any other means).

 

[Iliah Borg]

If you set EC to a negative value in order to get a darker picture, then the noise will not be affected.

Oh, but it will.

Yet another proof that you do not read.

Okay, I will put you on my Ignore User list as I find that very offensive.

Read again and come with another answer.

Nonetheless, I will respond yet again here. Yes, I read what you wrote, and I have done the experiment that shows that it is wrong. Your assertion (as I read it before, and as I still read it) is that there is no difference in the noise choosing a higher ISO and lowering brightness during Raw conversion or choosing a lower ISO and raising brightness during Raw conversion, and my answer to that remains the same.

That is not what I wrote above.

I wrote that if you use a negative EC in order to make the image darker, then you do not increase the noise.

And this is true.

It appears you need to re-read what you wrote right after that:

"But, if you also compensate in your editor to make the picture bright again, then you have increased noise, the same way as increasing ISO does."

My experiment showed that is wrong.

Finding a way to justify that ISO raise causes more noise is polishing a turd. That idea is fundamentally flawed.

 

[Roland Karlsson]

Nonetheless, higher ISO does not necessarily mean higher noise. It only happens when you put numerous qualifications to that statement (e.g., no use of EC or adjusting brightness by any other means).

Of course. Without constraints you can say nothing at all. About anything.

If you want to be able to say anything you have to very clearly define the conditions. So - let us see what we can say.

• If you have the lens cap on and take 10 second images with different ISO, then (for most cameras) the noise (in the RAW file) will increase with increasing ISO. Not for all though. For e.g. some Sigma cameras there is no variable amplifier before the A/D converter. In that case the noise will be constant. But most will have more noise for ISO 65,000 than ISO 100. NOTE - that te in camera JPEG will be very much more noisy for ISO 65,000 than 100, for all cameras.
• If you set your camera at auto exposure and increase the ISO (without changing EC) than the exposure will decrease, and the resulting image have the same brightness. Then the noise will increase. Maybe you need to change the ISO substantially in order to really see it.
• If you set your camera at a fixed ISO and decrease the exposure, then the image will become darker. But, the noise will not be affected at all. It will just be darker. Of course, if you compensate for the image being darker, and makes it bright again, then you will increase the noise.
• If you want to minimize noise, then the method Illia recommended is a good one. You choose the biggest aperture and the slowest exposure time you can accept, and then adjust the brightness of the image with ISO. This is the Tav mode on my Pentax. Which I sometimes use.

Now, most people do not specify so thoroughly what they mean. Maybe assuming that the reader makes reasonable assumptions. And often they do, or maybe do not care. But sometimes it goes wrong.

 

[Tony Beach]

Nonetheless, higher ISO does not necessarily mean higher noise. It only happens when you put numerous qualifications to that statement (e.g., no use of EC or adjusting brightness by any other means).

Of course. Without constraints you can say nothing at all. About anything.

Likewise if you put too many constraints (i.e., qualifications) on to something you are also not saying anything and are just arriving at your preconceived answer.

If you want to be able to say anything you have to very clearly define the conditions. So - let us see what we can say.

• If you have the lens cap on and take 10 second images with different ISO, then (for most cameras) the noise (in the RAW file) will increase with increasing ISO. Not for all though. For e.g. some Sigma cameras there is no variable amplifier before the A/D converter. In that case the noise will be constant. But most will have more noise for ISO 65,000 than ISO 100. NOTE - that te in camera JPEG will be very much more noisy for ISO 65,000 than 100, for all cameras.

As soon as I take the lens cap off though then I have light (that is, an EV), and with the lens cap on there is no SNR because there is no signal. Let's keep this discussion applicable to the real world, and not to taking photos of the inside of our lens cap.

• If you set your camera at auto exposure and increase the ISO (without changing EC) than the exposure will decrease, and the resulting image have the same brightness. Then the noise will increase. Maybe you need to change the ISO substantially in order to really see it.

Then it is the decrease in EV that is causing the noise, not the increase in ISO.

• If you set your camera at a fixed ISO and decrease the exposure, then the image will become darker. But, the noise will not be affected at all. It will just be darker. Of course, if you compensate for the image being darker, and makes it bright again, then you will increase the noise.

Which begs the question I asked and you actually answered (wrongly). Which approach increases noise more? The correct answer for most cameras is that it will be noisier if you set the ISO lower and compensate for that in development than it will be if you take the opposite approach; and in your best case scenario using a Sigma camera you can only claim that it will be the same (i.e., not better by lowering the ISO).

• If you want to minimize noise, then the method Illia recommended is a good one. You choose the biggest aperture and the slowest exposure time you can accept, and then adjust the brightness of the image with ISO. This is the Tav mode on my Pentax. Which I sometimes use.

Now, most people do not specify so thoroughly what they mean. Maybe assuming that the reader makes reasonable assumptions. And often they do, or maybe do not care. But sometimes it goes wrong.

Here's the thing. If you want to reduce the advice to what effect does raising or lowering the ISO have on noise, then the correct answer is that raising ISO will increase the SNR and thus lower the noise at any given EV.

 

[Roland Karlsson]

Finding a way to justify that ISO raise causes more noise is polishing a turd. That idea is fundamentally flawed.

The main reason for noise is lack of signal. I know that.

In principle you could set your camera at ISO 100 and then under expose when it is dark and boost the picture in your photo editor. Some Sigma cameras do that. And the major advantage is that you will get a fantastic protection against clipping.

And I know that there are some ISO-less cameras/sensors for which this works just fine.

Now - I do not think all cameras are ISO-less, so I do not think it applies to all cameras.

And - the opposite do not apply. You cannot set your camera at ISO 100,000 and take noise free images, if they not are pitch black. So, setting your camera to ISO 100,000 will absolutely generate lots of noise, if you want an image that is interesting to look at.

Moreover . ISO are two things. One is the setting of the exposure meter and the other is the internal amplifications set in the camera. So, if you want to state that changing ISO do not affect noise, you need to know which one of the two you are talking about.

Now, if you set the camera to auto exposure, then changing ISO changes both the exposure meter and the internal settings. And then, if you increase the ISO, than you also increase noise.

Or rather, in general. If you change the ISO of the exposure meter and then follows the advice from the exposure meter, then you will get a noisier result if you increase ISO. Simply because you will catch fewer photons.

So we are back

The main reason for noise is lack of signal.

 

[Roland Karlsson]

Here's the thing. If you want to reduce the advice to what effect does raising or lowering the ISO have on noise, then the correct answer is that raising ISO will increase the SNR and thus lower the noise at any given EV.

That is only true as long as you do not clip the signal.

The dynamic range at ISO 100,000 is extremely low.

I would challenge you to show me any low noise images taken at the max ISO of your camera. I would guess the image is mostly noise.

 

[Iliah Borg]

In principle you could set your camera at ISO 100 and then under expose

What is underexposure, if exposure is the same and it is independent of ISO? Exposure is about light, and light only.

You cannot set your camera at ISO 100,000 and take noise free images

If exposure is good enough, what is the source of the noise, ISO 100,000? Hm...

ISO are two things. One is the setting of the exposure meter and the other is the internal amplifications set in the camera.

ISO is not always changing the amplification in a camera. There is no rules here, not in ISO standard. Canon use the same ISO for different amplification settings in some recent cameras.

So, if you want to state that changing ISO do not affect noise

That's not what I want to say. Changing ISO reduces the noise or does nearly nothing to noise. It doesn't increase noise.

you need to know which one of the two you are talking about.

I think one always need to know what he is talking about. In this case, it is not even difficult.

Now, if you set the camera to auto exposure, then changing ISO changes both the exposure meter and the internal settings. And then, if you increase the ISO, than you also increase noise.

That's a little like saying when I close my eyes it's night.

Or rather, in general. If you change the ISO of the exposure meter and then follows the advice from the exposure meter, then you will get a noisier result if you increase ISO.

ISO last.

 

[Roland Karlsson]

You cannot set your camera at ISO 100,000 and take noise free images

If exposure is good enough, what is the source of the noise, ISO 100,000? Hm...

The problem is that if the exposure is good enough, the image is almost entirely clipped. Either by the analog amplifier before A/D converter or (for higher ISO) the digital amplification made before storing the image in 14 bits.

So - I challenge you to take your camera, use your highest ISO, and make low noise images.

I know of no one else that does it. Someone has to be first

 

[Tony Beach]

Finding a way to justify that ISO raise causes more noise is polishing a turd. That idea is fundamentally flawed.

The main reason for noise is lack of signal. I know that.

No, that's only half the equation. Noise is a low SNR.

In principle you could set your camera at ISO 100 and then under expose when it is dark and boost the picture in your photo editor. Some Sigma cameras do that. And the major advantage is that you will get a fantastic protection against clipping.

And I know that there are some ISO-less cameras/sensors for which this works just fine.

Now - I do not think all cameras are ISO-less, so I do not think it applies to all cameras.

It doesn't apply to most cameras, and some much less than others.

And - the opposite do not apply. You cannot set your camera at ISO 100,000 and take noise free images, if they not are pitch black. So, setting your camera to ISO 100,000 will absolutely generate lots of noise, if you want an image that is interesting to look at.

Moreover . ISO are two things. One is the setting of the exposure meter and the other is the internal amplifications set in the camera. So, if you want to state that changing ISO do not affect noise, you need to know which one of the two you are talking about.

Now, if you set the camera to auto exposure, then changing ISO changes both the exposure meter and the internal settings. And then, if you increase the ISO, than you also increase noise.

Back to fundamentals. ISO compensates for a low signal, which is why it is the inverse of Exposure Compensation (the former compensates after the exposure whereas the latter compensates before the exposure). You and the OP are confusing the effect with the cause and saying that the compensation for the lower EV is the cause of the noise when in fact is the other way around.

Or rather, in general. If you change the ISO of the exposure meter and then follows the advice from the exposure meter, then you will get a noisier result if you increase ISO. Simply because you will catch fewer photons.

I can also change the exposure meter by changing EC. Ergo, lowering EC causes noise (not).

So we are back

The main reason for noise is lack of signal.

Then how does reducing the amplification of the signal increase the SNR?

 

[Tony Beach]

Here's the thing. If you want to reduce the advice to what effect does raising or lowering the ISO have on noise, then the correct answer is that raising ISO will increase the SNR and thus lower the noise at any given EV.

That is only true as long as you do not clip the signal.

It's a separate question with a separate answer. Decreasing exposure decreases dynamic range. The way we see reduced dynamic range is in increasingly muddy shadows and eventually even in the mid-tones.

The dynamic range at ISO 100,000 is extremely low.

I would challenge you to show me any low noise images taken at the max ISO of your camera. I would guess the image is mostly noise.

What I showed you is that ISO 3200 is less noisy than ISO 100 using the same EV. Now you show me the opposite of that, short of that you have failed to prove me wrong about that.

 

[Iliah Borg]

You cannot set your camera at ISO 100,000 and take noise free images

If exposure is good enough, what is the source of the noise, ISO 100,000? Hm...

The problem is that if the exposure is good enough, the image is almost entirely clipped.

Why do I need to use ISO 100,000 then? It is either exposure is good, or the image is clipped.

I challenge you to take your camera, use your highest ISO, and make low noise images.

I did it with D1, 20 years ago. Noise was low enough to post here in 2003.

 

[Roland Karlsson]

The main reason for noise is lack of signal. I know that.

No, that's only half the equation. Noise is a low SNR.

Due to shot noise - that is much more than half the equation. And - I said the MAIN reason, not the WHOLE reason. I think you are nit picking.

Now - I do not think all cameras are ISO-less, so I do not think it applies to all cameras.

It doesn't apply to most cameras, and some much less than others.

OK

Back to fundamentals. ISO compensates for a low signal, which is why it is the inverse of Exposure Compensation (the former compensates after the exposure whereas the latter compensates before the exposure). You and the OP are confusing the effect with the cause and saying that the compensation for the lower EV is the cause of the noise when in fact is the other way around.

I have never said that ISO causes noise. I have only said that if you use auto exposure and increases ISO, you do increase noise. And you do. You just have to take one image at ISO 100 and one at ISO 64,000 and take a look. But, then the increased ISO only indirectly causes the noise bu reducing the exposure. And this is true.

The main reason for noise is lack of signal.

Then how does reducing the amplification of the signal increase the SNR?

By (in practice) giving room for the image above the noise! If you set your camera at ISO 100,000 then the noise will be moved up in the image stored in the file and there will be very little room left where to put the actual image, without clipping. So, then you have to move the image down into the noise.

That is why you cannot out your camera at ISO 100,000 and take noise free images.

 

[Roland Karlsson]

What I showed you is that ISO 3200 is less noisy than ISO 100 using the same EV. Now you show me the opposite of that, short of that you have failed to prove me wrong about that.

This is of course true. You can decrease noise by increasing ISO. By changing the ISO you have moved this part of the image from a very low value in the RAW file to a higher one. And, if the camera not is ISO-less, this improves the S/N, of course.

And this you can do as long as you do not clip any important parts of the image.

Note also that you could, for the ISO 100 image, increase the exposure to an higher EV. And then probably got a yet noise freer image.